Innovative Cognitive Stroke Detection Analysis using Deep Learning Methodology

Stroke is the second leading cause of death, and the World Health Organization defined stroke as ‘a rapidly progressive set of clinical symptoms characterized by central (or global), persistent brain dysfunction for more than 24 hours or causes death, either by intracranial or seizure; is the most common. Recently, artificial intelligence (AI) has shown great promise in medical imaging, in which computed tomography (CT) imaging was the primary method for determining whether a stroke is ischemic or hemorrhagic, and AI solutions are a potential tool that can be used for non-contrast head CT imaging for ICH diagnosis and therefore limit false negatives. To study and evaluate the current literature on stroke diagnosis in non-contrast computed tomography (NCCT) images and to test a deep learning model using NCCT images for stroke diagnosis. 2000 stroke brain window CT images were obtained online from the public database. The sample size determined was 186 images with an accuracy level of ±7% and a confidence level of 95%. The RESNET-18 model was used as a deep learning model to classify cerebral hemorrhage and their characteristics. Based on manual testing of the images in the AI model, the model achieved a high sensitivity and specificity of 70% for the training data set and 30% for the test data set. Out of 186 images, 155 images were ICH-positive and 31 images were ICH-negative (normal). The deep learning model achieved 86% overall accuracy for prediction of intracranial hemorrhage (ICH) and classified its subtypes with 86.2% sensitivity, 97.1% specificity, 14.1% error, and 88% precision. In 186 images, 178 images were correctly detected by the algorithm, and eight characters were mislabeled. Among the ICH subtypes, SAH was the best, and SDH was the least active subtype. The proposed method was able to correctly detect ICH and its subtypes, such as IVH, SDH, SAH, and EDH, with high accuracy (86%) and differentiate between ICH-negative (normal), which means it can help radiologists with their clinical assessment process. Our work confirms that, in clinical settings, AI solutions have the potential to support radiologists and act as a real-time clinical assistant for non-contrast head CT scans to exclude ICH and its types that can reduce waiting and detection times are more effective.